Modifying weights layer-by-layer with Levenberg-Marquardt backpropagation algorithm

Training feedforward networks, layer-by-layer, with the Levenberg-Marquardt backpropagation algorithm is presented in this paper. The Levenberg-Marqua rdt backpropagation technique has been noted as an efficient method for tra ining feedforward neural networks in terms of training accuracy, convergenc e properties and overall training time. We introduce a method to further im prove the computation and memory complexity of this algorithm by modifying the weights layer-by-layer. Four examples from the literature and from an e ngineering application are provided to demonstrate the ourperformance of th e technique over the general Levenberg-Marquardt backpropagation, which is based on adjusting all the weights simultaneously. These examples show that further improvement, in both the training time and convergence property, c an be obtained using the new approach.